feat: AgentChat streaming UX — answer_chunk typewriter, token/cost display, stop marker, session manage, date dividers

- Backend: token-level streaming via on_delta callback + asyncio.Queue bridge, answer_chunk SSE events; final event carries token_usage with cost_yuan; kimi pricing in cost_estimator
- Frontend: typewriter rendering of answer chunks; token/cost in message meta; abort marks message '已停止' and no longer triggers duplicate non-stream fallback; session pin/delete in dropdown (pinned first); date dividers across days

backend/app/agent_runtime/core.py

@@ -14,6 +14,7 @@ import hashlib
 import json
 import logging
 import time
+import asyncio
 from typing import Any, AsyncGenerator, Callable, Dict, List, Optional, Protocol, TypedDict
 
 from app.agent_runtime.schemas import (
@@ -271,13 +272,23 @@ class AgentRuntime:
 
     def _real_token_usage_payload(self) -> Dict[str, Any]:
         """构造随终态事件下发的真实用量 dict（形状含 total_tokens，供下游/前端与计费消费）。"""
-        return {
+        payload = {
             "prompt_tokens": self._real_usage["prompt_tokens"],
             "completion_tokens": self._real_usage["completion_tokens"],
             "total_tokens": self._real_usage["total_tokens"],
             "llm_calls": self._real_usage["llm_calls"],
             "model": self.config.llm.model,
         }
+        try:
+            from app.services.cost_estimator import estimate_cost_yuan
+            payload["cost_yuan"] = estimate_cost_yuan(
+                payload["model"] or "",
+                payload["prompt_tokens"],
+                payload["completion_tokens"],
+            )
+        except Exception:
+            pass
+        return payload
 
     def _attach_token_usage(self, result: AgentResult) -> AgentResult:
         """将 token 用量摘要附加到 AgentResult。预算启用则用其摘要；否则用真实累计用量。"""
@@ -978,15 +989,36 @@ class AgentRuntime:
             # think 事件：告知前端 Agent 正在思考（让 UI 即时反馈，避免假死感）
             yield {"type": "think", "content": "", "reasoning": None, "iteration": self.context.iteration}
 
-            # 调用 LLM
+            # 调用 LLM（token 级流式：delta 进队列，边下边 yield answer_chunk）
             try:
-                response = await llm.chat(
+                delta_q: asyncio.Queue = asyncio.Queue()
+
+                def _on_delta(text: str) -> None:
+                    try:
+                        delta_q.put_nowait(text)
+                    except Exception:
+                        pass
+
+                chat_task = asyncio.create_task(llm.chat(
                     messages=messages,
                     tools=tool_schemas if has_tools and self.context.iteration == 1 else
                                (tool_schemas if has_tools else None),
                     iteration=self.context.iteration,
                     on_completion=_llm_callback,
-                )
+                    on_delta=_on_delta,
+                ))
+                while True:
+                    try:
+                        piece = await asyncio.wait_for(delta_q.get(), timeout=0.2)
+                        yield {"type": "answer_chunk", "content": piece,
+                               "iteration": self.context.iteration}
+                    except asyncio.TimeoutError:
+                        if chat_task.done():
+                            break
+                while not delta_q.empty():
+                    yield {"type": "answer_chunk", "content": delta_q.get_nowait(),
+                           "iteration": self.context.iteration}
+                response = chat_task.result()
             except Exception as e:
                 err_str = str(e)
                 logger.error("LLM 调用失败 (iteration=%s): %s", self.context.iteration, err_str)
@@ -1803,8 +1835,10 @@ class _LLMClient:
         tools: Optional[List[Dict[str, Any]]] = None,
         iteration: int = 1,
         on_completion: Optional[Callable[[Dict[str, Any]], Any]] = None,
+        on_delta: Optional[Callable[[str], None]] = None,
     ) -> Any:
-        """调用 LLM，主模型失败时自动切换 fallback_llm 重试。"""
+        """调用 LLM，主模型失败时自动切换 fallback_llm 重试。
+        on_delta 提供时启用 token 级流式（仅内容增量，工具调用仍会重组后返回）。"""
         from openai import AsyncOpenAI
         from app.core.config import settings
 
@@ -1822,7 +1856,7 @@ class _LLMClient:
         return await self._do_chat(
             api_key=api_key, base_url=base_url, model=self._config.model,
             messages=messages, tools=tools, iteration=iteration,
-            on_completion=on_completion,
+            on_completion=on_completion, on_delta=on_delta,
         )
 
     async def _do_chat(
@@ -1835,6 +1869,7 @@ class _LLMClient:
         iteration: int = 1,
         on_completion: Optional[Callable[[Dict[str, Any]], Any]] = None,
         _is_fallback: bool = False,
+        on_delta: Optional[Callable[[str], None]] = None,
     ) -> Any:
         from openai import AsyncOpenAI
         from app.core.config import settings
@@ -1883,10 +1918,22 @@ class _LLMClient:
                     tool_calls = None
                 return _CachedMsg()
 
+        # token 级流式：要求增量回调时走 stream=True，末 chunk 带 usage
+        if on_delta:
+            kwargs["stream"] = True
+            kwargs["stream_options"] = {"include_usage": True}
+
         start_time = time.perf_counter()
         last_error = None
+        message = None
+        usage = None
         try:
-            response = await client.chat.completions.create(**kwargs)
+            if on_delta:
+                message, usage = await self._stream_chat(client, kwargs, on_delta)
+            else:
+                response = await client.chat.completions.create(**kwargs)
+                message = response.choices[0].message
+                usage = getattr(response, "usage", None)
         except Exception as e:
             last_error = e
 
@@ -1914,6 +1961,7 @@ class _LLMClient:
                             iteration=iteration,
                             on_completion=on_completion,
                             _is_fallback=_is_fallback,
+                            on_delta=on_delta,
                         )
                 except Exception as ce:
                     logger.error("ReactiveCompact 失败: %s", ce)
@@ -1953,12 +2001,11 @@ class _LLMClient:
                         model=fb_model,
                         messages=messages, tools=tools,
                         iteration=iteration, on_completion=on_completion,
-                        _is_fallback=True,
+                        _is_fallback=True, on_delta=on_delta,
                     )
             raise
 
         latency_ms = int((time.perf_counter() - start_time) * 1000)
-        message = response.choices[0].message
 
         # 缓存写入（仅不用工具时）
         if self._config.cache_enabled and not tools and message.content:
@@ -1966,7 +2013,6 @@ class _LLMClient:
             await _llm_cache_set(ck, message.content, self._config.cache_ttl_ms)
 
         # 提取 token 用量
-        usage = getattr(response, "usage", None)
         prompt_tokens = usage.prompt_tokens if usage else 0
         completion_tokens = usage.completion_tokens if usage else 0
         total_tokens = usage.total_tokens if usage else 0
@@ -1985,3 +2031,61 @@ class _LLMClient:
             })
 
         return message
+
+    @staticmethod
+    async def _stream_chat(client: Any, kwargs: Dict[str, Any], on_delta: Callable[[str], None]) -> Any:
+        """消费 OpenAI 兼容流式响应：逐 delta 回调 on_delta，重组为 message 对象返回。
+
+        返回 (message, usage)：message 为 SimpleNamespace（content/tool_calls/
+        reasoning_content，duck-type 兼容 SDK Message），usage 来自开启
+        stream_options.include_usage 后的末 chunk。
+        """
+        from types import SimpleNamespace
+
+        stream = await client.chat.completions.create(**kwargs)
+        content_parts: List[str] = []
+        reasoning_parts: List[str] = []
+        tool_slots: Dict[int, Dict[str, Any]] = {}
+        usage = None
+
+        async for chunk in stream:
+            if getattr(chunk, "usage", None):
+                usage = chunk.usage
+            if not chunk.choices:
+                continue
+            delta = chunk.choices[0].delta
+            piece = getattr(delta, "content", None)
+            if piece:
+                content_parts.append(piece)
+                try:
+                    on_delta(piece)
+                except Exception:
+                    pass
+            reasoning_piece = getattr(delta, "reasoning_content", None)
+            if reasoning_piece:
+                reasoning_parts.append(reasoning_piece)
+            for tc in (getattr(delta, "tool_calls", None) or []):
+                slot = tool_slots.setdefault(tc.index, {"id": None, "name": "", "arguments": ""})
+                if tc.id:
+                    slot["id"] = tc.id
+                fn = getattr(tc, "function", None)
+                if fn is not None:
+                    if getattr(fn, "name", None):
+                        slot["name"] += fn.name
+                    if getattr(fn, "arguments", None):
+                        slot["arguments"] += fn.arguments
+
+        tool_calls = [
+            SimpleNamespace(
+                id=s["id"], type="function",
+                function=SimpleNamespace(name=s["name"], arguments=s["arguments"]),
+            )
+            for _, s in sorted(tool_slots.items())
+        ] or None
+
+        message = SimpleNamespace(
+            content="".join(content_parts) if content_parts else None,
+            tool_calls=tool_calls,
+            reasoning_content="".join(reasoning_parts) if reasoning_parts else None,
+        )
+        return message, usage

backend/app/services/cost_estimator.py

@@ -23,6 +23,8 @@ MODEL_PRICING: Dict[str, Tuple[float, float]] = {
     "claude-3-haiku": (0.25, 1.25),
     "claude-3.5-sonnet": (3.00, 15.00),
     "claude-3.5-haiku": (1.00, 5.00),
+    "kimi-k3": (3.00, 15.00),
+    "kimi-k2": (0.56, 2.22),
 }
 
 # 缓存未命中模型的默认定价

frontend/src/components/ChatMessageBubble.vue

@@ -32,7 +32,9 @@
       <!-- Meta line -->
       <div class="msg-meta">
         <span class="msg-time">{{ formatTime(timestamp) }}</span>
+        <el-tag v-if="status === 'stopped'" size="small" type="warning" effect="plain">已停止</el-tag>
         <span v-if="iterations" class="msg-iterations">{{ iterations }} 步 · {{ toolCallsMade }} 次工具调用</span>
+        <span v-if="tokenUsageText" class="msg-token-usage">{{ tokenUsageText }}</span>
         <span class="msg-actions">
           <slot name="actions" />
         </span>
@@ -54,6 +56,7 @@ const props = defineProps<{
   attachments?: MessageAttachment[]
   iterations?: number
   toolCallsMade?: number
+  tokenUsage?: any
 }>()
 
 defineEmits<{
@@ -62,6 +65,14 @@ defineEmits<{
 
 const attachmentsClass = computed(() => (props.role === 'user' ? 'is-user' : 'is-agent'))
 
+const tokenUsageText = computed(() => {
+  const u = props.tokenUsage
+  if (!u || !u.total_tokens) return ''
+  let text = `${u.total_tokens} tokens`
+  if (u.cost_yuan !== undefined && u.cost_yuan !== null) text += ` · ¥${u.cost_yuan}`
+  return text
+})
+
 function formatTime(ts: number): string {
   const d = new Date(ts)
   return d.toLocaleTimeString('zh-CN', { hour: '2-digit', minute: '2-digit' })
@@ -194,6 +205,7 @@ function formatTime(ts: number): string {
   flex-wrap: wrap;
 }
 .msg-iterations { color: var(--el-color-info); }
+.msg-token-usage { color: var(--el-text-color-placeholder); }
 .msg-actions {
   margin-left: auto;
   display: flex;

frontend/src/views/AgentChat.vue

@@ -33,14 +33,33 @@
               <span style="color: var(--el-color-primary); font-weight: 500">+ 新对话</span>
             </el-option>
             <el-option
-              v-for="s in sessions"
+              v-for="s in sortedSessions"
               :key="s.session_id"
               :label="s.title || '未命名对话'"
               :value="s.session_id"
             >
-              <div style="display:flex;flex-direction:column;width:100%">
-                <span class="session-option-title">{{ s.title || '未命名对话' }}</span>
-                <span class="session-option-meta">{{ s.message_count }} 条消息 · {{ relativeTimeText(s.updated_at) }}</span>
+              <div class="session-option-row">
+                <div style="display:flex;flex-direction:column;flex:1;min-width:0">
+                  <span class="session-option-title">
+                    <el-icon v-if="s.is_pinned" :size="12" style="color:var(--el-color-warning);vertical-align:-1px"><Top /></el-icon>
+                    {{ s.title || '未命名对话' }}
+                  </span>
+                  <span class="session-option-meta">{{ s.message_count }} 条消息 · {{ relativeTimeText(s.updated_at) }}</span>
+                </div>
+                <span class="session-option-actions">
+                  <el-icon
+                    :size="14"
+                    :style="{ color: s.is_pinned ? 'var(--el-color-warning)' : 'var(--el-text-color-placeholder)' }"
+                    :title="s.is_pinned ? '取消置顶' : '置顶'"
+                    @click.stop.prevent="togglePinSession(s)"
+                  ><Top /></el-icon>
+                  <el-icon
+                    :size="14"
+                    style="color:var(--el-color-danger)"
+                    title="删除会话"
+                    @click.stop.prevent="deleteSession(s)"
+                  ><Delete /></el-icon>
+                </span>
               </div>
             </el-option>
           </el-select>
@@ -170,24 +189,26 @@
         </template>
       </div>
 
-      <ChatMessageBubble
-        v-for="(msg, i) in displayMessages"
-        :key="i"
-        :role="msg.role"
-        :content-html="renderMarkdown(msg.orchestrateResult?.final_answer || msg.content)"
-        :timestamp="msg.timestamp"
-        :status="msg.status"
-        :iterations="msg.iterations"
-        :tool-calls-made="msg.tool_calls_made"
-      >
-        <template v-if="msg.orchestrateResult" #extra>
+      <template v-for="item in displayItems" :key="item.key">
+        <div v-if="item._divider" class="date-divider"><span class="date-divider-text">{{ item.label }}</span></div>
+        <ChatMessageBubble
+          v-else
+          :role="item.msg.role"
+          :content-html="renderMarkdown(item.msg.orchestrateResult?.final_answer || item.msg.content)"
+          :timestamp="item.msg.timestamp"
+          :status="item.msg.status"
+          :iterations="item.msg.iterations"
+          :tool-calls-made="item.msg.tool_calls_made"
+          :token-usage="item.msg.token_usage"
+        >
+        <template v-if="item.msg.orchestrateResult" #extra>
           <div class="orchestrate-result">
             <div class="orch-header">
-              <el-tag size="small" type="info">{{ msg.orchestrateResult.mode }}</el-tag>
-              <span class="orch-agent-count">{{ msg.orchestrateResult.steps.length }} 个 Agent</span>
+              <el-tag size="small" type="info">{{ item.msg.orchestrateResult.mode }}</el-tag>
+              <span class="orch-agent-count">{{ item.msg.orchestrateResult.steps.length }} 个 Agent</span>
             </div>
             <div class="orch-steps">
-              <div v-for="(step, si) in msg.orchestrateResult.steps" :key="si" class="orch-step" :class="{ expanded: step._open }">
+              <div v-for="(step, si) in item.msg.orchestrateResult.steps" :key="si" class="orch-step" :class="{ expanded: step._open }">
                 <div class="orch-step-header" @click="step._open = !step._open">
                   <el-icon><CaretRight :style="{ transform: step._open ? 'rotate(90deg)' : '' }" /></el-icon>
                   <el-tag size="small" :type="step.error ? 'danger' : 'success'" round>{{ step.agent_name }}</el-tag>
@@ -201,21 +222,21 @@
           </div>
         </template>
 
-        <template v-if="!msg.orchestrateResult" #extra>
-          <div v-if="msg.tool_calls && msg.tool_calls.length > 0" class="tool-calls">
-            <div class="tool-calls-header"><el-icon><Tools /></el-icon> 工具调用 ({{ msg.tool_calls.length }})</div>
-            <div v-for="(tc, j) in msg.tool_calls" :key="j" class="tool-call-item">
+        <template v-if="!item.msg.orchestrateResult" #extra>
+          <div v-if="item.msg.tool_calls && item.msg.tool_calls.length > 0" class="tool-calls">
+            <div class="tool-calls-header"><el-icon><Tools /></el-icon> 工具调用 ({{ item.msg.tool_calls.length }})</div>
+            <div v-for="(tc, j) in item.msg.tool_calls" :key="j" class="tool-call-item">
               <span class="tool-name">{{ tc.function?.name || '?' }}</span>
               <el-tag size="small" type="info">{{ safeParseArgCount(tc.function?.arguments) }} 个参数</el-tag>
             </div>
           </div>
-          <div v-if="msg.steps && msg.steps.length > 0" class="thinking-trace">
-            <div class="trace-header" @click="toggleTrace(msg)">
-              <el-icon><CaretRight :style="{ transform: msg._traceOpen ? 'rotate(90deg)' : '' }" /></el-icon>
-              <span>思考链 ({{ msg.steps.length }} 步)</span>
+          <div v-if="item.msg.steps && item.msg.steps.length > 0" class="thinking-trace">
+            <div class="trace-header" @click="toggleTrace(item.msg)">
+              <el-icon><CaretRight :style="{ transform: item.msg._traceOpen ? 'rotate(90deg)' : '' }" /></el-icon>
+              <span>思考链 ({{ item.msg.steps.length }} 步)</span>
             </div>
-            <div v-show="msg._traceOpen" class="trace-steps">
-              <div v-for="(step, si) in msg.steps" :key="si" class="trace-step" :class="'step-' + step.type">
+            <div v-show="item.msg._traceOpen" class="trace-steps">
+              <div v-for="(step, si) in item.msg.steps" :key="si" class="trace-step" :class="'step-' + step.type">
                 <div class="step-icon">
                   <el-icon v-if="step.type === 'think'"><ChatDotSquare /></el-icon>
                   <el-icon v-else-if="step.type === 'tool_result'"><Tools /></el-icon>
@@ -247,16 +268,17 @@
         </template>
 
         <template #actions>
-          <el-button v-if="msg.role === 'assistant' && msg.content" link size="small" @click="isSpeaking ? stopSpeak() : speakMessage(msg.orchestrateResult?.final_answer || msg.content)" :title="isSpeaking ? '停止朗读' : '朗读'">
+          <el-button v-if="item.msg.role === 'assistant' && item.msg.content" link size="small" @click="isSpeaking ? stopSpeak() : speakMessage(item.msg.orchestrateResult?.final_answer || item.msg.content)" :title="isSpeaking ? '停止朗读' : '朗读'">
             <el-icon><Headset v-if="!isSpeaking" /><VideoPause v-else /></el-icon>
           </el-button>
-          <el-button v-if="msg.role === 'user'" link size="small" @click="editMessage(i)" title="编辑"><el-icon><Edit /></el-icon></el-button>
-          <el-button v-if="msg.role === 'assistant'" link size="small" @click="copyMessage(msg)" title="复制"><el-icon><DocumentCopy /></el-icon></el-button>
-          <el-button link size="small" @click="createBranch(i)" title="从此分支"><el-icon><Share /></el-icon></el-button>
-          <el-button v-if="msg.role === 'assistant' && !loading" link :type="msg.status === 'error' ? 'danger' : undefined" size="small" @click="retryMessage(i)" :title="msg.status === 'error' ? '重试' : '重新生成'"><el-icon><Refresh /></el-icon></el-button>
-          <el-button link size="small" @click="deleteMessage(i)" title="删除"><el-icon style="color:var(--el-color-danger)"><Delete /></el-icon></el-button>
+          <el-button v-if="item.msg.role === 'user'" link size="small" @click="editMessage(item.idx)" title="编辑"><el-icon><Edit /></el-icon></el-button>
+          <el-button v-if="item.msg.role === 'assistant'" link size="small" @click="copyMessage(item.msg)" title="复制"><el-icon><DocumentCopy /></el-icon></el-button>
+          <el-button link size="small" @click="createBranch(item.idx)" title="从此分支"><el-icon><Share /></el-icon></el-button>
+          <el-button v-if="item.msg.role === 'assistant' && !loading" link :type="item.msg.status === 'error' ? 'danger' : undefined" size="small" @click="retryMessage(item.idx)" :title="item.msg.status === 'error' ? '重试' : '重新生成'"><el-icon><Refresh /></el-icon></el-button>
+          <el-button link size="small" @click="deleteMessage(item.idx)" title="删除"><el-icon style="color:var(--el-color-danger)"><Delete /></el-icon></el-button>
         </template>
-      </ChatMessageBubble>
+        </ChatMessageBubble>
+      </template>
 
       <div v-if="loading && !streamingActive" class="message assistant">
         <div class="message-avatar"><el-avatar :size="36" icon="Promotion" /></div>
@@ -449,7 +471,7 @@
 import { ref, computed, watch, onMounted, onUnmounted, nextTick } from 'vue'
 import { useRoute } from 'vue-router'
 import { ElMessage, ElMessageBox } from 'element-plus'
-import { ChatLineSquare, UserFilled, Promotion, Tools, CaretRight, ChatDotSquare, Select, DocumentCopy, Refresh, VideoPause, Edit, Delete, Headset, Setting, Search, Download, Share, MagicStick } from '@element-plus/icons-vue'
+import { ChatLineSquare, UserFilled, Promotion, Tools, CaretRight, ChatDotSquare, Select, DocumentCopy, Refresh, VideoPause, Edit, Delete, Headset, Setting, Search, Download, Share, MagicStick, Top } from '@element-plus/icons-vue'
 import MainLayout from '@/components/MainLayout.vue'
 import ChatMessageBubble from '@/components/ChatMessageBubble.vue'
 import ChatInputArea from '@/components/ChatInputArea.vue'
@@ -481,6 +503,7 @@ interface OrchestrateResult {
 interface ChatMessage {
   role: 'user' | 'assistant'; content: string; tool_calls?: any[]; timestamp: number
   iterations?: number; tool_calls_made?: number; status?: string; steps?: AgentStep[]
+  token_usage?: any
   _traceOpen?: boolean; orchestrateResult?: OrchestrateResult
 }
 interface OrchestrateAgentForm {
@@ -612,6 +635,33 @@ const currentAgentKey = computed(() => {
 })
 const displayMessages = computed(() => messages.value[currentAgentKey.value] || [])
 
+// 日期分割线：消息列表中跨天时插入分隔项
+interface DisplayItem {
+  _divider?: boolean; label?: string; key: string
+  msg?: ChatMessage; idx?: number
+}
+function _dayStart(ts: number): number { const d = new Date(ts); d.setHours(0, 0, 0, 0); return d.getTime() }
+function dayLabelText(ts: number): string {
+  const days = Math.round((_dayStart(Date.now()) - _dayStart(ts)) / 86400000)
+  if (days <= 0) return '今天'
+  if (days === 1) return '昨天'
+  if (days < 7) return `${days} 天前`
+  return new Date(ts).toLocaleDateString('zh-CN', { year: 'numeric', month: 'long', day: 'numeric' })
+}
+const displayItems = computed<DisplayItem[]>(() => {
+  const items: DisplayItem[] = []
+  let lastDay = -1
+  displayMessages.value.forEach((msg, idx) => {
+    const day = _dayStart(msg.timestamp || Date.now())
+    if (day !== lastDay) {
+      lastDay = day
+      items.push({ _divider: true, label: dayLabelText(msg.timestamp || Date.now()), key: `divider-${day}` })
+    }
+    items.push({ key: `msg-${idx}-${msg.timestamp}`, msg, idx })
+  })
+  return items
+})
+
 // 编排模式
 const chatMode = ref<'single' | 'orchestrate'>('single')
 const orchestrateMode = ref('debate')
@@ -891,6 +941,38 @@ function newSession() {
   nextTick(scrollToBottom)
 }
 
+// 置顶优先，其余按更新时间倒序
+const sortedSessions = computed(() =>
+  [...sessions.value].sort((a, b) => {
+    if (!!a.is_pinned !== !!b.is_pinned) return a.is_pinned ? -1 : 1
+    return new Date(b.updated_at || 0).getTime() - new Date(a.updated_at || 0).getTime()
+  })
+)
+
+async function togglePinSession(s: SessionInfo) {
+  try {
+    await api.patch(`/api/v1/agent-chat/${currentAgentId.value}/sessions/${s.session_id}`, { is_pinned: !s.is_pinned })
+    s.is_pinned = !s.is_pinned
+    ElMessage.success(s.is_pinned ? '已置顶' : '已取消置顶')
+  } catch { ElMessage.error('操作失败') }
+}
+
+async function deleteSession(s: SessionInfo) {
+  try {
+    await ElMessageBox.confirm(
+      `确定删除会话「${s.title || '未命名对话'}」吗？该操作不可恢复。`,
+      '删除会话',
+      { type: 'warning', confirmButtonText: '删除', cancelButtonText: '取消' }
+    )
+  } catch { return }
+  try {
+    await api.delete(`/api/v1/agent-chat/${currentAgentId.value}/sessions/${s.session_id}`)
+    sessions.value = sessions.value.filter(x => x.session_id !== s.session_id)
+    if (currentSessionId.value === s.session_id) newSession()
+    ElMessage.success('会话已删除')
+  } catch { ElMessage.error('删除失败') }
+}
+
 // ── 预设问题：LLM 按 Agent 描述生成，localStorage 缓存（Agent 更新后自动失效）──
 const presetQuestions = ref<string[]>([])
 const presetLoading = ref(false)
@@ -1044,7 +1126,11 @@ async function sendMessage() {
                 const data = JSON.parse(dataStr)
                 if (!receivedFirstEvent) { receivedFirstEvent = true; streamingActive.value = true }
                 if (eventType === 'think') {
+                  currentMsg.content = ''
                   currentMsg.steps!.push({ iteration: data.iteration, type: 'think', content: data.content || '思考中...', reasoning: data.reasoning, tool_name: data.tool_names?.[0] })
+                } else if (eventType === 'answer_chunk') {
+                  currentMsg.content = (currentMsg.content || '') + (data.content || '')
+                  scrollToBottom()
                 } else if (eventType === 'tool_call') {
                   currentMsg.steps!.push({ iteration: data.iteration, type: 'tool_call', content: `调用工具: ${data.name}`, tool_name: data.name, tool_input: data.input })
                 } else if (eventType === 'tool_result') {
@@ -1055,6 +1141,7 @@ async function sendMessage() {
                 } else if (eventType === 'final') {
                   clearTimeout(streamTimeout)
                   currentMsg.content = data.content || ''; currentMsg.iterations = data.iterations_used || 0; currentMsg.tool_calls_made = data.tool_calls_made || 0
+                  if (data.token_usage) currentMsg.token_usage = data.token_usage
                   if (data.session_id) { sessionId.value[key] = data.session_id; if (!currentSessionId.value) { currentSessionId.value = data.session_id; loadSessions() } }
                   streamingActive.value = false; loading.value = false
                 } else if (eventType === 'error') {
@@ -1066,7 +1153,19 @@ async function sendMessage() {
           }
           clearTimeout(streamTimeout)
         }
-      } catch { clearTimeout(streamTimeout); usedStreaming = false; streamingActive.value = false }
+      } catch {
+        clearTimeout(streamTimeout); streamingActive.value = false
+        if (abortController.value?.signal.aborted) {
+          // 用户主动停止：标记「已停止」，阻止非流式 fallback 重复请求
+          if (placeholderIdx >= 0 && placeholderIdx < messages.value[key].length) {
+            const m = messages.value[key][placeholderIdx]
+            if (m.content || (m.steps && m.steps.length)) { m.status = 'stopped' } else { messages.value[key].splice(placeholderIdx, 1) }
+          }
+          loading.value = false; saveState(); nextTick(scrollToBottom)
+          return
+        }
+        usedStreaming = false
+      }
 
       if (!usedStreaming) {
         if (placeholderIdx >= 0 && placeholderIdx < messages.value[key].length) messages.value[key].splice(placeholderIdx, 1)
@@ -1231,6 +1330,26 @@ const hasMoreHistory = ref(false)
 
 /* Session selector */
 .session-option-title { font-size: 13px; line-height: 1.4; overflow: hidden; text-overflow: ellipsis; white-space: nowrap; max-width: 260px; }
+.session-option-row { display: flex; align-items: center; width: 100%; gap: 8px; }
+.session-option-actions { display: flex; gap: 8px; align-items: center; flex-shrink: 0; }
+.session-option-actions .el-icon { cursor: pointer; padding: 2px; }
+.session-option-actions .el-icon:hover { transform: scale(1.15); }
+.date-divider {
+  display: flex; align-items: center; justify-content: center;
+  margin: 16px 0 8px; width: 100%;
+  position: relative;
+}
+.date-divider::before {
+  content: ''; position: absolute; left: 0; right: 0; top: 50%;
+  border-top: 1px solid var(--el-border-color-lighter);
+}
+.date-divider-text {
+  position: relative; z-index: 1;
+  background: var(--el-bg-color);
+  padding: 2px 12px;
+  font-size: 12px;
+  color: var(--el-text-color-placeholder);
+}
 .session-option-meta { font-size: 11px; color: var(--el-text-color-placeholder); }
 
 /* Orchestrate */